Abstract
Purpose :
To provide a free user friendly downloadable software for PCs, as well as an online version for any platform, capable of enhancing retinal images when blurred by cataracts or corneal opacities and thereby improve diagnostics.
Methods :
We employed a multiple light scattering model inspired by the generalized Gaussian distribution for particles suspended in the atmosphera to derive an eye convolution kernel that serves as an effective tool for deconvolving eye fundus images exhibiting diferent degrees of severity. The forward scattering parameter of the model related to the severity of the opacitities is the only adjustable parameter for the three color channels. An intuitive user friendly interface has been develoved for uploading eye fundus images, adjusting the scattering parameter and saving the improved image for posterior analysis.
Results :
The software has been texted with over 150 images sourced from both a public dataset and those provided by CHUS. In the case of immature cataracts, the software exhibits a significant enhancement, notably improving the visualization of retinal vessels, especially the arterial ones, along with the choroidal fundus and macular area. Drusen, if present, are clearly discernible. In the context of mature cataracts, the improvement is slightly less pronounced compared to the previous scenario; however, retinal vessels and choroidal fundus remain more observable. In the case of hypermature cataracts, the image processing allows in some of the analized pictures for a partial detection of retinal vessels, and the optical disc becomes perceptible, but other details are not yet observable.
Conclusions :
In this work we introduce a new optical filter to approximate multiple scattering of light rays within eyes with opacities and deconvolve eye fundus images afected by eye opacities. The user-friendly software implementation with the eye convolution kernel facilitates enhanced images, making it a valuable tool for the diagnosis of retinal pathologies.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.